Re: [PATCH] generic/371: run write(2) and fallocate(2) in parallel

[Eryu Guan <eguan@redhat.com>]

On Thu, Jul 21, 2016 at 03:30:25PM +0800, Wang Xiaoguang wrote:
> Currently in btrfs, there is something wrong with fallocate(2)'s data
> space reservation, it'll temporarily occupy more data space thant it
> really needs, which in turn will impact other operations' data request.
> 
> In this test case, it runs write(2) and fallocate(2) in parallel and the
> total needed data space for these two operations don't exceed whole fs
> free data space, to see whether we will get any unexpected ENOSPC error.
> 
> Signed-off-by: Wang Xiaoguang <wangxg.fnst@cn.fujitsu.com>
> ---
>  tests/generic/371     | 119 ++++++++++++++++++++++++++++++++++++++++++++++++++
>  tests/generic/371.out |   2 +
>  tests/generic/group   |   1 +
>  3 files changed, 122 insertions(+)
>  create mode 100755 tests/generic/371
>  create mode 100644 tests/generic/371.out
> 
> diff --git a/tests/generic/371 b/tests/generic/371
> new file mode 100755
> index 0000000..b85327a
> --- /dev/null
> +++ b/tests/generic/371
> @@ -0,0 +1,119 @@
> +#! /bin/bash
> +# FS QA Test 371
> +#
> +# Run write(2) and fallocate(2) in parallel and the total needed data space
> +# for these operations don't exceed whole fs free data space, to see whether
> +# we will get any unexpected ENOSPC error.
> +#
> +#-----------------------------------------------------------------------
> +# Copyright (c) 2016 Fujitsu.  All Rights Reserved.
> +#
> +# This program is free software; you can redistribute it and/or
> +# modify it under the terms of the GNU General Public License as
> +# published by the Free Software Foundation.
> +#
> +# This program is distributed in the hope that it would be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program; if not, write the Free Software Foundation,
> +# Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
> +#-----------------------------------------------------------------------
> +#
> +
> +seq=`basename $0`
> +seqres=$RESULT_DIR/$seq
> +echo "QA output created by $seq"
> +
> +here=`pwd`
> +tmp=/tmp/$$
> +status=1	# failure is the default!
> +trap "_cleanup; exit \$status" 0 1 2 3 15
> +
> +_cleanup()
> +{
> +	cd /
> +	rm -f $tmp.*
> +}
> +
> +# get standard environment, filters and checks
> +. ./common/rc
> +. ./common/filter
> +
> +# remove previous $seqres.full before test
> +rm -f $seqres.full
> +
> +# Modify as appropriate.
> +_supported_fs generic
> +_supported_os Linux
> +_require_scratch

Need '_require_xfs_io_command "falloc"', otherwise test fails on ext3/2,
because they don't support fallocate(2).

> +
> +_scratch_mkfs_sized $((256 * 1024 * 1024)) >> $seqres.full 2>&1
> +_scratch_mount
> +
> +testfile1=$SCRATCH_MNT/testfile1
> +testfile2=$SCRATCH_MNT/testfile2
> +
> +write_work()
> +{
> +	rm -f $testfile1
> +	while [ 1 ]; do
> +		$XFS_IO_PROG -f -c "pwrite 0 80M" $testfile1 >>$seqres.full 2>&1
> +		grep "No space left on device" $seqres.full >/dev/null
> +		if [ $? -eq 0 ]; then
> +			echo "unexpected ENOSPC error occurs"
> +			exit 1
> +		fi

No need to grep for error message, just redirect stdout to /dev/null and
any error will appear in stderr, which will break golden image.

And $seqres.full is where we dump logs for debug purpose, tests should
dump output they need to somewhere like $tmp.<suffix>

> +		rm -f $testfile1
> +	done
> +}
> +
> +fallocate_work()
> +{
> +	rm -f  $testfile2
> +	while [ 1 ]; do
> +		$XFS_IO_PROG -f -c "falloc 0 80M" $testfile2 >>$seqres.full 2>&1
> +		grep "No space left on device" $seqres.full >/dev/null
> +		if [ $? -eq 0 ]; then
> +			echo "unexpected ENOSPC error occurs"
> +			exit 1
> +		fi
> +	rm -f $testfile2
> +	done
> +}
> +
> +run_time=$((180 * $TIME_FACTOR))

180s is too long time, I can reproduce it in around 10s on my test vm,
just loop for 100 times for each operation (pwrite and falloc)

> +write_work &
> +write_work_pid=$!
> +fallocate_work &
> +fallocate_work_pid=$!
> +
> +for ((elapsed_time = 0; elapsed_time < run_time; elapsed_time += 5)); do
> +	kill -0 $write_work_pid >/dev/null 2>&1
> +	if [ $? -ne 0 ]; then
> +		kill $fallocate_work_pid >/dev/null 2>&1
> +		break
> +	fi
> +
> +	kill -0 $fallocate_work_pid >/dev/null 2>&1
> +	if [ $? -ne 0 ]; then
> +		kill $write_work_pid >/dev/null 2>&1
> +		break
> +	fi
> +	sleep 5
> +done
> +
> +kill $fallocate_work_pid >/dev/null 2>&1
> +kill $write_work_pid >/dev/null 2>&1
> +wait
> +
> +# wait un-finished xfs_io
> +while ps aux | grep "xfs_io" | grep -qv grep; do
> +	sleep 1
> +done

And this seems unnecessarily complicated

So I'd write it as:

echo "Silence is golden"
for ((i=0; i<100; i++)); do
        $XFS_IO_PROG -fc "pwrite 0 80M" $testfile1 >/dev/null
        rm -f $testfile1
done &
pids=$!
for ((i=0; i<100; i++)); do
        $XFS_IO_PROG -fc "falloc 0 80M" $testfile2 >/dev/null
        rm -f $testfile2
done &
pids="$pids $!"

wait $pids
status=0
exit

btrfs failed due to extra "pwrite64: No space left on device" output from
xfs_io, and ext4 and xfs passed the test within 15s for me.

> +
> +echo "Silence is golden"
> +status=0
> +exit
> diff --git a/tests/generic/371.out b/tests/generic/371.out
> new file mode 100644
> index 0000000..22ec8a2
> --- /dev/null
> +++ b/tests/generic/371.out
> @@ -0,0 +1,2 @@
> +QA output created by 371
> +Silence is golden
> diff --git a/tests/generic/group b/tests/generic/group
> index 97ecb65..3d4a802 100644
> --- a/tests/generic/group
> +++ b/tests/generic/group
> @@ -373,3 +373,4 @@
>  368 auto quick richacl
>  369 auto quick richacl
>  370 auto quick richacl
> +371 auto enospc prealloc stress

So we can add 'quick' group and remove 'stress'.

Thanks,
Eryu